February 24, 2008

To:  Southern European and Latin American social policy workshop participants
From:  John Stephens
Re:  Comparative figures on poverty, inequality, and social spending

I have assembled some figures on inequality, poverty, and social spending in the seven countries we will be discussing at the workshop.  I wish I could leave it at that, but since the data are not necessarily comparable across regions I have to make some commentary on them

Inequality and poverty data:  I have included data from three sources:  LIS (mainly OECD countries, http://www.lisproject.org), SEDLAC (Latin America and the Caribbean, http://www.depeco.econo.unlp.edu.ar/cedlas/sedlac/), and WIID (world wide, http://www.wider.unu.edu/research/Database/en_GB/database/).  The best in terms of being most comparable, highest quality and accessibility of the micro-data is LIS.  Unfortunately, there are no LIS data for Portugal and the only Latin American country for which there are LIS data is Mexico.  Janet Gornick, the LIS director, has told Evelyne and me that LIS will be adding Guatemala, Uruguay, Peru, Colombia, Brazil, and possibly Argentina in the coming year, which is great but does not help us now.  The new SEDLAC database also offers a high level of comparability.  The problem here is that it is Latin America and the Caribbean only and the equivalence scales (household size adjustments) that it uses are not the same as the LIS data, so it is not directly comparable.  The WIID data contain all seven of our countries.  The problem in this case is that the data for individual country years are not comparable to each other and they are of varying quality.  However, the information WIID provides about the surveys on which the inequality measures are based makes it possible to produce a dataset with comparable data.  I have done that for our seven countries in the sheet labeled “household per capita,” which is the household size adjustment used in this table.  

A broader set of the WIID data is in the sheet marked “WIID2b.”  This is already just a subset of all of the WIID observations for these seven countries, as I have deleted the observations which were not “all” on population coverage and age coverage and those which were based on consumption, expenditure, or earnings rather than income.  I also retained only those with “household” on income sharing unit and “person” on unit of analysis.  There are differences in the equivalence scales used.  Let me remark on this because this does make a difference for the degree of inequality.  The most common equivalence scales are household per capita, the WIID convention, and square root of household per capita, the LIS convention.  There are good reasons why LIS and WIID differ on this.  In developing countries, especially low income developing countries, a huge proportion of household income goes to food, so it makes sense to divide the household income by the number of persons in the household.  By contrast, this is not true of OECD countries.  In these counties, there are large household economies of scales, particularly in housing, which is the reason why LIS uses the square root of the household size as the divisor.  SEDLAC presents the data for the per capita divisor but argues this is not really appropriate for Latin America.  The SEDLAC researchers present a number of alternate adjustments which are somewhere between per capita and the square root of per capita.  I have included one in the sheet marked “LA gini”.  The formula for this one is (#Adults+ .3#Kids under 6+.5#Kids 6-14).75.   

For comparisons of inequality within the regions between countries and through time, I recommend using the SEDLAC data (sheet “LA gini”) and the LIS data.  For comparisons between the regions, one must use the WIID data in the sheet “household per capita”.  You will note that the data for Uruguay and Argentina are for urban areas only.  In a World Bank paper, Londońo and Székely (1997) argue that one can safely use urban data as a very good proxy for national data for these two countries because they are very urban, and the evidence we do have indicates that inclusion of rural areas would not change the degree of inequality.  You will also note that the income definition varies among the Latin American cases, with some observations having disposable income, some gross income, and some have no information on whether they were gross or net income.  SEDLAC researchers find that these differences have little effect on the inequality measures, presumably because income taxes are low and not very progressive in Latin America.  This is also what we found in our analysis using the WIID data (Huber et al., ASR, December 2006).

The poverty data are even more variable between the regions.  Here we have a conceptual difference.  The convention in comparative studies of OECD countries is to measure poverty in relative terms.  The LIS standard is a poverty line defined as 50% of the median household income while the European Union standard is 60% of the median.  I include both for several different groups in the sheet with the LIS data.  In studies of developing countries, the convention is to measure poverty in absolute terms, with World Bank’s $2 per day (actually $2.15) being the most common measure.  Based on its collection of household surveys, SEDLAC calculates it own version of the World Bank measure, which makes adjustments for the difference in cost of a basic food basket in rural and urban areas and for household size.  It is included in the sheet, “LA poverty”.  The United Nations’ Economic Commission for Latin America and the Caribbean (ECLAC) argues that the World Bank measure does not reflect differences in cost of a basic food basket across countries and between rural and urban areas.  The ECLAC measure does this.  I include it in the LA poverty sheet.  Finally, the LA poverty sheet includes the SEDLAC calculation of the 50% of median income poverty line (the LIS relative poverty line).

Several of the sheets contain data on social spending in Latin America and Southern Europe.  The Southern Europe data are from the SOCX database and the OECD’s Education at Glance.  The Latin America data are from our database.  I will post our codebook on the workshop website which documents the sources for the spending data.  These data are not comparable across the two regions but they do give one an idea of the order of magnitude of the differences in social spending.  In particular, one can see that social security and welfare spending in Uruguay is far higher than in the other Latin American countries and comparable to the three Southern European countries.